Phase 2: Implement Execution Environment Abstraction (v0.3.0)

This commit implements Phase 2 of the CHORUS Task Execution Engine development plan,
providing a comprehensive execution environment abstraction layer with Docker
container sandboxing support.

## New Features

### Core Sandbox Interface
- Comprehensive ExecutionSandbox interface with isolated task execution
- Support for command execution, file I/O, environment management
- Resource usage monitoring and sandbox lifecycle management
- Standardized error handling with SandboxError types and categories

### Docker Container Sandbox Implementation
- Full Docker API integration with secure container creation
- Transparent repository mounting with configurable read/write access
- Advanced security policies with capability dropping and privilege controls
- Comprehensive resource limits (CPU, memory, disk, processes, file handles)
- Support for tmpfs mounts, masked paths, and read-only bind mounts
- Container lifecycle management with proper cleanup and health monitoring

### Security & Resource Management
- Configurable security policies with SELinux, AppArmor, and Seccomp support
- Fine-grained capability management with secure defaults
- Network isolation options with configurable DNS and proxy settings
- Resource monitoring with real-time CPU, memory, and network usage tracking
- Comprehensive ulimits configuration for process and file handle limits

### Repository Integration
- Seamless repository mounting from local paths to container workspaces
- Git configuration support with user credentials and global settings
- File inclusion/exclusion patterns for selective repository access
- Configurable permissions and ownership for mounted repositories

### Testing Infrastructure
- Comprehensive test suite with 60+ test cases covering all functionality
- Docker integration tests with Alpine Linux containers (skipped in short mode)
- Mock sandbox implementation for unit testing without Docker dependencies
- Security policy validation tests with read-only filesystem enforcement
- Resource usage monitoring and cleanup verification tests

## Technical Details

### Dependencies Added
- github.com/docker/docker v28.4.0+incompatible - Docker API client
- github.com/docker/go-connections v0.6.0 - Docker connection utilities
- github.com/docker/go-units v0.5.0 - Docker units and formatting
- Associated Docker API dependencies for complete container management

### Architecture
- Interface-driven design enabling multiple sandbox implementations
- Comprehensive configuration structures for all sandbox aspects
- Resource usage tracking with detailed metrics collection
- Error handling with retryable error classification
- Proper cleanup and resource management throughout sandbox lifecycle

### Compatibility
- Maintains backward compatibility with existing CHORUS architecture
- Designed for future integration with Phase 3 Core Task Execution Engine
- Extensible design supporting additional sandbox implementations (VM, process)

This Phase 2 implementation provides the foundation for secure, isolated task
execution that will be integrated with the AI model providers from Phase 1
in the upcoming Phase 3 development.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

vendor/github.com/go-logr/logr/sloghandler.go

@@ -0,0 +1,192 @@
+//go:build go1.21
+// +build go1.21
+
+/*
+Copyright 2023 The logr Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package logr
+
+import (
+	"context"
+	"log/slog"
+)
+
+type slogHandler struct {
+	// May be nil, in which case all logs get discarded.
+	sink LogSink
+	// Non-nil if sink is non-nil and implements SlogSink.
+	slogSink SlogSink
+
+	// groupPrefix collects values from WithGroup calls. It gets added as
+	// prefix to value keys when handling a log record.
+	groupPrefix string
+
+	// levelBias can be set when constructing the handler to influence the
+	// slog.Level of log records. A positive levelBias reduces the
+	// slog.Level value. slog has no API to influence this value after the
+	// handler got created, so it can only be set indirectly through
+	// Logger.V.
+	levelBias slog.Level
+}
+
+var _ slog.Handler = &slogHandler{}
+
+// groupSeparator is used to concatenate WithGroup names and attribute keys.
+const groupSeparator = "."
+
+// GetLevel is used for black box unit testing.
+func (l *slogHandler) GetLevel() slog.Level {
+	return l.levelBias
+}
+
+func (l *slogHandler) Enabled(_ context.Context, level slog.Level) bool {
+	return l.sink != nil && (level >= slog.LevelError || l.sink.Enabled(l.levelFromSlog(level)))
+}
+
+func (l *slogHandler) Handle(ctx context.Context, record slog.Record) error {
+	if l.slogSink != nil {
+		// Only adjust verbosity level of log entries < slog.LevelError.
+		if record.Level < slog.LevelError {
+			record.Level -= l.levelBias
+		}
+		return l.slogSink.Handle(ctx, record)
+	}
+
+	// No need to check for nil sink here because Handle will only be called
+	// when Enabled returned true.
+
+	kvList := make([]any, 0, 2*record.NumAttrs())
+	record.Attrs(func(attr slog.Attr) bool {
+		kvList = attrToKVs(attr, l.groupPrefix, kvList)
+		return true
+	})
+	if record.Level >= slog.LevelError {
+		l.sinkWithCallDepth().Error(nil, record.Message, kvList...)
+	} else {
+		level := l.levelFromSlog(record.Level)
+		l.sinkWithCallDepth().Info(level, record.Message, kvList...)
+	}
+	return nil
+}
+
+// sinkWithCallDepth adjusts the stack unwinding so that when Error or Info
+// are called by Handle, code in slog gets skipped.
+//
+// This offset currently (Go 1.21.0) works for calls through
+// slog.New(ToSlogHandler(...)).  There's no guarantee that the call
+// chain won't change. Wrapping the handler will also break unwinding. It's
+// still better than not adjusting at all....
+//
+// This cannot be done when constructing the handler because FromSlogHandler needs
+// access to the original sink without this adjustment. A second copy would
+// work, but then WithAttrs would have to be called for both of them.
+func (l *slogHandler) sinkWithCallDepth() LogSink {
+	if sink, ok := l.sink.(CallDepthLogSink); ok {
+		return sink.WithCallDepth(2)
+	}
+	return l.sink
+}
+
+func (l *slogHandler) WithAttrs(attrs []slog.Attr) slog.Handler {
+	if l.sink == nil || len(attrs) == 0 {
+		return l
+	}
+
+	clone := *l
+	if l.slogSink != nil {
+		clone.slogSink = l.slogSink.WithAttrs(attrs)
+		clone.sink = clone.slogSink
+	} else {
+		kvList := make([]any, 0, 2*len(attrs))
+		for _, attr := range attrs {
+			kvList = attrToKVs(attr, l.groupPrefix, kvList)
+		}
+		clone.sink = l.sink.WithValues(kvList...)
+	}
+	return &clone
+}
+
+func (l *slogHandler) WithGroup(name string) slog.Handler {
+	if l.sink == nil {
+		return l
+	}
+	if name == "" {
+		// slog says to inline empty groups
+		return l
+	}
+	clone := *l
+	if l.slogSink != nil {
+		clone.slogSink = l.slogSink.WithGroup(name)
+		clone.sink = clone.slogSink
+	} else {
+		clone.groupPrefix = addPrefix(clone.groupPrefix, name)
+	}
+	return &clone
+}
+
+// attrToKVs appends a slog.Attr to a logr-style kvList.  It handle slog Groups
+// and other details of slog.
+func attrToKVs(attr slog.Attr, groupPrefix string, kvList []any) []any {
+	attrVal := attr.Value.Resolve()
+	if attrVal.Kind() == slog.KindGroup {
+		groupVal := attrVal.Group()
+		grpKVs := make([]any, 0, 2*len(groupVal))
+		prefix := groupPrefix
+		if attr.Key != "" {
+			prefix = addPrefix(groupPrefix, attr.Key)
+		}
+		for _, attr := range groupVal {
+			grpKVs = attrToKVs(attr, prefix, grpKVs)
+		}
+		kvList = append(kvList, grpKVs...)
+	} else if attr.Key != "" {
+		kvList = append(kvList, addPrefix(groupPrefix, attr.Key), attrVal.Any())
+	}
+
+	return kvList
+}
+
+func addPrefix(prefix, name string) string {
+	if prefix == "" {
+		return name
+	}
+	if name == "" {
+		return prefix
+	}
+	return prefix + groupSeparator + name
+}
+
+// levelFromSlog adjusts the level by the logger's verbosity and negates it.
+// It ensures that the result is >= 0. This is necessary because the result is
+// passed to a LogSink and that API did not historically document whether
+// levels could be negative or what that meant.
+//
+// Some example usage:
+//
+//	logrV0 := getMyLogger()
+//	logrV2 := logrV0.V(2)
+//	slogV2 := slog.New(logr.ToSlogHandler(logrV2))
+//	slogV2.Debug("msg") // =~ logrV2.V(4) =~ logrV0.V(6)
+//	slogV2.Info("msg")  // =~  logrV2.V(0) =~ logrV0.V(2)
+//	slogv2.Warn("msg")  // =~ logrV2.V(-4) =~ logrV0.V(0)
+func (l *slogHandler) levelFromSlog(level slog.Level) int {
+	result := -level
+	result += l.levelBias // in case the original Logger had a V level
+	if result < 0 {
+		result = 0 // because LogSink doesn't expect negative V levels
+	}
+	return int(result)
+}